Disentangling inertial waves from eddy turbulence in a forced rotating-turbulence experiment Antoine Campagne,1 Basile Gallet,2 Frederic Moisy,1 and Pierre-Philippe Cortet1 1Laboratoire FAST, CNRS, Universite Paris-Sud, 91405 Orsay, France 1Lahoratoire SPHYNX, Service de Physique de 1’Etat Condense, CEA Saclay, CNRS UMR 3680, 91191 Gif-sur-Yvette, France (Received 13 January 2015; published 23 April 2015) We present a spatiotemporal analysis of a statistically stationary rotating-turbulence experiment, aiming to extract a signature of inertial waves and to determine the scales and frequencies at which they can be detected. The analysis uses two-point spatial correlations of the temporal Fourier transform of velocity fields obtained from time-resolved stereoscopic particle image velocimetry measurements in the rotating frame. We quantify the degree of anisotropy of turbulence as a function of frequency and spatial scale. We show that this space-time-dependent anisotropy is well described by the dispersion relation of linear inertial waves at large scale, while smaller scales are dominated by the sweeping of the waves by fluid motion at larger scales. This sweeping effect is mostly due to the low-trequency quasi-two-dimensional component ot the turbulent flow, a prominent feature of our experiment that is not accounted for by wave-turbulence theory. These results question the relevance of this theory for rotating turbulence at the moderate Rossby numbers accessible in laboratory experiments, which are relevant to most geophysical and astrophysical flows. DOl: 10.1103/PhysRevE.91.043016

PACS number(s): 47.27.- i , 47.32.Ef, 47.35.- i

I. INTRODUCTION

The energy content of turbulence is usually characterized by the energy distribution among spatial scales, either in physical or in Fourier space. For rotating, stratified, or magnetohydrodynamic turbulence [1], waves can propagate and coexist with classical eddies and coherent structures, which advocates for a spatiotemporal description of such flows. While temporal fluctuations are usually slaved to the spatial ones via sweeping effects in classical turbulence [2,3], they are expected to be governed by the dispersion relation of the waves for time scales much smaller than the eddy turn over time. The latter regime is the subject of wave-turbulence theory, in which the assumption of weak nonlinear coupling between waves allows one to predict scaling laws for the spatial energy spectrum [4,5]. It is a matter of debate whether wave-turbulence theory (also known as weak-turbulence theory) is a good candidate to describe rotating turbulence in the rapidly rotating limit. Solutions to the linearized rotating Euler equation can be decomposed into inertial waves, which satisfy the anisotropic dispersion relation

where £2 is the rotation rate and k\\ the component of the wave vector k along the rotation axis (referred to as the vertical axis by convention) [6], Accordingly, only fluid motions at frequencies a smaller than the Coriolis frequency 2£2 correspond to wave propagation. Fluid motions of weak amplitude and slowly varying in time (a < 2£2) can be described in terms of waves with nearly horizontal wave vectors: They tend to be two-dimensional three-component (2D3C), invariant along the rotation axis, a result known as the Taylor-Proudman theorem. The trend towards two-dimensionality is a landmark in rotating turbulence, observed in both experiments and numer ical simulations [1,7-11], It originates from the modification of the nonlinear interactions by the Coriolis force, which 1539-3755/2015/91 (4)/043016(10)

yields preferential energy transfers towards modes with almost horizontal wave vectors [12-15], In the frequency domain, this corresponds to the generation of slow dynamics compared to the characteristic frequency at which energy is supplied to the system. These anisotropic energy transfers can be accounted for in terms of resonant and nearly resonant triadic interactions of inertial waves [16-19], A major feature of rotating turbulence is the emergence of inverse energy transfers in the horizontal plane [13,19-26], Inverse transfers between 3D fast wave modes and the 2D slow vortex mode, mediated by near-resonant triadic interactions, are allowed at finite Rossby number only [22,23,27-29]. The 2D mode is therefore fed either from the coupling with the 3D modes at finite Rossby number or from direct energy input by the forcing. One naturally expects the energy within this 2D mode to undergo an inverse energy cascade, similar to that of classical (nonrotating) 2D turbulence [30,31], Such coexistence between 2D and 3D flows is rele vant to most experiments and numerical simulations and cannot be accounted for by wave-turbulence theory, which describes the direct energy cascade arising from resonant triadic interactions of 3D wave modes only [32,33], This theory therefore provides only a partial description of rotating turbulence in realistic systems and careful experimental and numerical studies remain necessary to assess its range of validity. Laboratory experiments differ from most numerical and theoretical studies by the presence of rigid horizontal bound aries, where the rotating flow achieves no-slip conditions through Ekman layers [34], In a laminar Ekman layer, the balance between the viscous and Coriolis forces leads to a J v / S l . The belief is that, boundary layer thickness % provided the experimental tank is deep enough, the bulk turbulent flow away from the top and bottom boundaries should resemble the one obtained in the ideal 3D periodic or stress-free domains considered in most numerical and theoretical studies. Closer to the horizontal walls, the boundary layers induce Ekman friction that is not taken into account by most numerical studies.

043016-1

©2015 American Physical Society

PHYSICAL REVIEW E 91,043016 (2015)

CAMPAGNE, GALLET, MOISY, AND CORTET

In the laboratory, the energy dissipation of rotating turbulent flows originates from three main contributions: bulk viscous dissipation of 3D flow structures, bulk viscous dissipation of quasi-2D flow structures (somewhat similar to the bulk energy dissipation of 2D turbulence), and dissipation through Ekman friction on the horizontal boundaries. In spite of the importance of the 2D mode in most geophysical and laboratory flows, the 3D fluctuations still play a crucial role in the dynamics of rotating turbulence, because they are much more efficient at dissipating energy. This key feature is illustrated in Fig. 2, from data obtained in the present experiment (setup sketched in Fig. 1; see Sec. Ill for details): We decompose the turbulent velocity field into a vertically averaged 2D flow and a vertically dependent 3D remainder and show the corresponding energies and energy dissipation rates as a function of global rotation. For maximum rotation, although the 3D component contains a small fraction of the total kinetic energy, its dissipation rate is as large as that of the vertically invariant 2D component. Moreover, both of these dissipations are larger than an estimate of the frictional losses due to laminar Ekman layers (see Sec. III). An accurate description of the 3D structures of the flow is therefore essential to characterize the energy fluxes in rotating turbulence at moderate Rossby number. A primary goal in this direction is to determine the range of scales and frequencies for which 3D fluctuations follow the inertial-wave dispersion relation. This requires a full spatiotemporal analysis, which is very demanding in general for wave-turbulence systems: The accessible range of scales is usually limited in experiments, whereas long integration times are prohibitive in numerical simulations. The case of rotating turbulence is particularly delicate because of the specific form of the dispersion relation (1): The frequency is not related to the wave number, as in conventional isotropic wave systems such as surface waves [35] or elastic waves [36,37], but to the wave-vector orientation only. The recent studies of Clark di Leoni et al. [38] and Yarom and Sharon [39] constitute important steps forward in this respect. Using numerical simulation of rotating turbulence forced at large scale, Clark di Leoni et al. [38] observe a clear concentration of energy along the dispersion relation of inertial waves and provide a detailed analysis of the various time scales of the system. They observe a wave-dominated regime at large scale and a sweeping-dominated regime at small scale (see also Ref. [40]). In the experiment of Yarom and Sharon [39] the forcing consists of a random set of sources and sinks at the bottom of a rotating water tank. They measure 3D2C velocity fields using a scanning particle image velocimetry (PIV) technique and observe also good agreement with the inertial-wave dispersion relation. In both Refs. [38] and [39], the inertial waves are observed at scales smaller than the injection scale, suggesting that they are fed by forward energy transfers, which is consistent with the predictions of wave-turbulence theory. The aim of the present paper is to further analyze experi mentally the range of spatiotemporal scales at which inertial waves can be detected in rotating turbulence. Stationary rotat ing turbulence is produced by a set of vortex dipole generators that continuously inject turbulent fluctuations towards the center of a rotating water tank where measurements are

Laser

200 cm

0

co

FIG. 1. (Color online) Experimental setup: (a) side view and (b) top view. An arena of ten pairs of flaps forces a turbulent flow in the central region of a water tank mounted on a rotating turntable. A laser sheet illuminates a vertical slice through a horizontal glass lid covering the fluid. Two-dimensional three-component velocity measurements are performed using stereoscopic particle image velocimetry in a vertical square domain of size Ax x Az = 14 x 14 cm2, shown as a dashed square in (a).

performed. We showed in Ref. [26] that this configuration generates a double energy cascade at large rotation rate: an inverse cascade of horizontal energy and a direct cascade of vertical energy, which behaves approximately as a passive scalar advected by the horizontal flow. Here we perform a detailed spatiotemporal analysis using two-point spatial correlations of the temporal Fourier modes computed from time-resolved 2D3C velocity fields measured by stereoscopic PIV in a vertical plane. We observe that, at large scales and frequencies, the spatiotemporal anisotropy of the energy distribution is well described by the dispersion relation of inertial waves, whereas smaller scales are dominated by the sweeping of the waves by the energetic large-scale flow.

043016-2

DISENTANGLING INERTIAL WAVES FROM EDDY . ..

PHYSICAL REVIEW E 91, 043016 (2015)

II. EXPERIMENTAL SETUP

The experimental setup, sketched in Fig. 1, is similar to the one described in Rets. [26,41] and is only briefly described here. It consists of a 125 x 125 x 65 cm3 glass tank, filled with H — 50 cm of water and mounted on a 2-m-diam rotating platform that rotates at a rate S2 in the range 0.21-1.68 rad s ^ 1 (2-16 rpm). Turbulence is produced in the rotating frame by a set of ten vertical vortex dipole generators organized as a circular arena of 85 cm diameter around the center of the water tank. This forcing device was initially designed to generate turbulence in stratified fluids and is described in detail in Refs. [42,43]. Each generator consists of a pair of vertical flaps, 60 cm high and L f = 10 cm long, alternatively closing rapidly and opening slowly in a cyclic motion of period To = 2jt/ ctq = 8.5 s. The closing stage is achieved with the flaps rotating at an angular velocity ay = 0.092 rad s-i and a random phase shift is applied between the generators. In the laminar regime, a single pair of flaps generates vortex dipoles with core vorticity Wf. Additional PIV measurements close to a vortex dipole generator indicate that this core vorticity is governed by the vorticity in the viscous boundary layer of the flap u>f ~ a f Lf / S , where Of L f is the flap velocity and & the viscous boundary layer thickness. In the present experiment, the vortex dipoles are unstable and the closing of the flaps therefore produces small-scale 3D turbulent fluctuations that are advected towards the center of the arena by the remaining large-scale dipolar structure. The turbulent Reynolds number, computed from the rms velocity and the horizontal integral scale, is about 400 in the center of the flow and the turbulent Rossby number covers the range 0.30-0.07 for f2 = 2-16 rpm [26]. We measure the three components of the velocity field u = « ,eA+ n ve v + uzez (with e; oriented vertically, along the rotation axis) in a vertical square domain of size A r x A z = 1 4 x l 4 cm2 located at the center of the circular arena at mid-depth, using a stereoscopic PIV system [44,45] embarked on the rotating platform. These 2D3C velocity fields are sampled on a grid of 80 x 80 vectors with a spatial resolution of 1.75 mm. Two acquisition sets are recorded for each rotation rate Q: one set of 10 000 fields at 0.35 Hz and one set of 1000 fields at 1.5 Hz. The combination of these two time series results in a temporal spectral range of three decades.

III. 2D VS 3D FLOW COMPONENTS

In the present experiment, energy is primarily injected in the 2D mode (vertically invariant), but the instabilities in the vicinity of the flaps rapidly feed 3D fluctuations that are advected in the central region. Energy transfers between the 2D and 3D flow components, which vanish in the weak-turbulence limit (Ro -> 0), are allowed in our system because of the moderate value of the Rossby number. It is therefore of interest to quantify the energy contained in the 2D and 3D components of the flow. We estimate the vertically averaged 2D flow as the average of the velocity field over the vertical extent Az of the PIV field, i U2D

r Az

= — / Az Jo

U{x,z)dz,

(2 )

fl (rpm ) FIG. 2. (Color online) (a) Energy and (b) energy dissipation rate per unit mass for the 2D and 3D modes as a function of the rotation rate Q. In both figures, the first data points (shown with arrows, at arbitrary abscissas) correspond to the nonrotating case Q = 0.

and the remaining z-dependent 3D flow as u3D = u —u2d We compute the energy per unit mass of these two flow com ponents as (u2d)/2 and (u2D)/2, with the overline denoting the temporal average and angular brackets the spatial average over the PIV field. They are plotted in Fig. 2(a) as a function of the rotation rate Q. Because of the limited height of the PIV field, the 2D flow estimated from Eq. (2) unavoidably contains 3D fluctuations associated with vertical scales larger than Az, so the measured 2D energy may overestimate the true one. Figure 2(a) shows that without rotation the 2D and 3D components of the flow have comparable energy. With rotation, the 2D energy increases with £2, following approximately the power law !IT/3 [41], whereas the 3D energy remains approximately constant and represents only 5% of the total energy at the largest rotation rate. Although most of the energy is contained in the 2D flow component for fi ^ 0, a significant fraction of the dissipation still arises from the 3D fluctuations. Assuming axisymmetry, we compute an estimate of the energy dissipation rate e = v((3w,/3x7)2) from the six terms of the

043016-3

PHYSICAL REVIEW E 91 , 043016 (2015)

CAMPAGNE, GALLET, MOISY, AND CORTET velocity gradient tensor accessible in the 3C2D m easurem ents,

This dissipation rate, com puted for both U2D and U3D, is shown in Fig. 2(b). Since the derivatives are obtained from finite differences at the sm allest resolved scale, the com puted dissipation underestim ates the true one (the PIV resolution is 1.75 mm w hile the K olmogorov scale is o f order o f 0.6 mm [41]). However, we expect the m easured evolution o f e with £2 to reflect the true one. We first com pare these bulk energy dissipation rates to an estim ate o f frictional losses due to lam inar Ekm an layers eEk ~ o jjj.

n

- \ / v £ 2 ^ f \ w here U±rm&is the root-m ean-square n

horizontal velocity. This estim ate ranges from 8 x 10~9 m 2 s~3 for £2 = 2 rpm to 1 x 10 -7 m2 s-3 for £2 = 16 rpm; it is sm aller than the bulk energy dissipation o f both the 2D and 3D parts o f the turbulent flow, by a factor o f 10 for slow rotation and 4 for rapid rotation. A detailed experim ental characterization o f these Ekm an layers w ould be necessary to validate the assum ption o f lam inar layers, but it is beyond the scope of the present study. We now com pare the bulk energy dissipation rates in the 2D and 3D parts o f the turbulent flow. Remarkably, w hile the 3D fluctuations represent a small fraction o f the total energy, they account for a large fraction o f the dissipation at all rotation rates. It is therefore o f interest to investigate these 3D modes and to determ ine to w hat extent they can be described in term s o f inertial waves. IV. TEMPORAL ANALYSIS We now focus on the tem poral dynam ics o f the velocity field, w hich w e characterize through the energy distribution o f turbulent fluctuations as a function o f angular frequency a . T his tem poral energy spectrum is defined as Ajr

E(cr) = — r ~ l are therefore damped, with rv = *Jv/o a viscous cutoff. This viscous cutoff is of the order of 1 to 10 mm for the normalized frequencies a* = cr/2Q. in the range [10 1] considered in Fig. 5. However, since viscous damping allects the wave amplitude without modifying the wave-vector components, its should not affect the anisotropy. We therefore focus in the following on the sweeping effect. B. Sweeping effect Sweeping corresponds to the advection of the waves by the large-scale flow, which leads to a modification of their apparent frequency. An inertial wave propagating in a time-independent uniform flow U has a Doppler-shifted frequency a = at + k • U,

(10)

where ay is the intrinsic frequency given by (1) and a is the frequency at which the wave is detected in the frame of the rotating tank. In our experiment, the energetic large-scale 2D flow may be thought of locally as a uniform sweeping flow U that evolves slowly in time, inducing a scrambling of the waves’ spatiotemporal signature. An order of magnitude of the typical Doppler shift can be estimated by where (/inns is the root-mean-square horizontal velocity. A key difference between Eqs. (10) and (1) is that the frequency a now depends on the magnitude of k. with small-scale waves more affected by sweeping. For an ensemble of inertial waves with axisymmetric wavevector statistics, the intrinsic frequency ay can be related to the anisotropy through Eq. (9). Substituting the corresponding expression into (10) and estimating the Doppler-shift term on dimensional grounds, we obtain 2^

t/lrm s

V 1+ 2A -2

GL

a

’

(ID

where C is a constant of order unity. This indicates that the parameter

2£2rj_ N

=

---------------;----------

(

12)

UXrms \ ! 1 + 2 / A FIG. 7. Anisotropy factor A (8) as a function of the normalized frequency cr* = a/2Sl at different rotation rates (same symbols as in Fig. 3), for three horizontal scales (a) rL = 9, (b) r± = 18, and (c) Gl = 50 mm. The solid line represents the inviscid inertial-wave prediction Aw (9).

horizontal scales and large rotation rate. For such large scales (G l — 50 mm), the anisotropy factor is no longer accessible for a* < 0 .1 because it corresponds to much larger than the height of the PIV field. On the other hand, at smaller horizontal scale the prediction (9) fails, with small frequencies more isotropic than predicted by the inertial-wave argument. Because of the moderate Reynolds and Rossby numbers of the present experiment, two effects may be considered to explain why large scales follow the inertial-wave prediction whereas small scales do not: viscous damping and sweeping of small scales by the velocity at larger scales. Viscosity introduces an imaginary term iu |k |2 in the dispersion relation

should be a unique function of the sweeping parameter 5 = t/Lrms/o'GL- Here, N corresponds approximately to the intrinsic frequency of inertial waves rescaled by the advective time gl/ ( 2j_rms, whereas 5 is the observed period of the waves, rescaled by the same advective time. Figure 8 confirms this picture: The data for different values of £2, r l, and a collapse onto a master curve N = /( 5 ) . This collapse indicates that sweeping is indeed responsible for the departure from the inertial-wave prediction at small frequencies and/or small scales. The expected asymptotic behavior for a small sweeping parameter is A ~ 1/5, which corresponds to the prediction (9) for an axisymmetric ensemble of nonswept inertial waves. The data are in quantitative agreement with this small-5 prediction, shown as a dashed line in Fig. 8. For large 5, Eq. (11) indicates that N should asymptotically approach a constant value N ~ C, which again is compatible with the data. The master curve in Fig. 8 has the following simple interpretation: High-frequency or large-scale waves are hardly

043016-7

PHYSICAL REVIEW E 91, 043016 (2015)

CAMPAGNE, GALLET, MOISY, AND CORTET

FIG. 8. Rescaled intrinsic frequency N (12) as a function of the sweeping parameter S. The symbols indicate the different rotation rates and are the same as in Fig. 3. The dashed line N ~ 1/S shows the low-5 prediction for (nonswept) ensembles of inertial waves [Eq. (12)]. For S < 1 one sees the signature of the dispersion relation (1), while for S 2> 1 one detects swept inertial waves.

em bedded in a turbulent 2D flow. T his is a form idable task in general because the 2D flow is space and tim e dependent: In the discussion o f our data, we sim plified the problem by assum ing that the 2D flow is at much larger scales and slower frequencies than the waves, therefore including it as a sim ple D oppler-shift term in the dispersion relation. We conclude w ith a discussion on the dim ensionality of the forcing. In the present experim ent, the flow is driven by vertically invariant flaps: Such a quasi-2D forcing device enhances two dim ensionalization and the resulting sw eeping of the 3D flow structures. N evertheless, accum ulation o f energy in the 2D m ode is a robust feature o f rotating turbulence, w hich takes place for arbitrary forcing geom etry, even if the forcing does not input energy directly into the 2D mode. A careful and extensive num erical study o f this issue has been recently reported for the fully 3D Taylor-G reen forcing [47]: For rapid global rotation and low viscosity, energy accum ulates in the 2D m ode until the Rossby num ber based on the velocity of this 2D flow is o f order unity. If these findings are confirmed, the sw eeping o f the m ost energetic 3D structures w ould be an inevitable outcom e o f this accum ulation o f 2D energy. ACKNOWLEDGMENTS

affected by sw eeping. The D oppler-shift term is negligible com pared to their intrinsic frequency and their location in a space-tim e energy distribution is given by the dispersion relation (1). This is the low -S behavior in Fig. 8 . By contrast, w hen focusing on low frequencies o or small scales in the fram e o f the tank, one m easures the inertial waves w ith intrinsic frequency 07 = 0 , but one also detects many waves w ith 07 ^ 0 that are D oppler shifted back to frequency 0 by the advective term in (10). The anisotropy m easured at low frequency 0 therefore results from strongly sw ept inertial w aves with various intrinsic frequencies and the inform ation from the dispersion relation ( 1) is lost in the space-tim e correlation. The lim it 5 » 1 corresponds to frequencies 0 that are m uch low er than the inverse advective time. In this 0 —> 0 lim it, one detects m ostly waves w ith 0 , » 0 that are D oppler shifted by the horizontal flow in such a way that they are alm ost steady in the fram e o f the tank: This is a 0 -independent regim e that corresponds to the large-S plateau in Fig. 8.

We acknow ledge R Augier, P. Billant, and J.-M . C h o m azfo r kindly providing the flap apparatus and A. A ubertin, L. Auffray, C. Borget, and R. Pidoux for their experim ental help. This w ork was supported by the A N R G rant No. 2 0 1 1-BS04-006-01 “O N LIT U R ” . This w ork is supported by “Investissem ents d ’A venir” LabEx PALM (A N R -10-LA B X -0039-PA LM ). F.M. acknow ledges the Institut Universitaire de France. APPENDIX: ANISOTROPY FACTOR FOR A STATISTICALLY AXISYMMETRIC DISTRIBUTION OF INERTIAL WAVES We com pute the anisotropy factor A for an ensem ble of independent plane inertial waves, w ith axisym m etric wavevector statistics. The tem poral Fourier transform o f the velocity field reads u (x , 0 ) = / a ( k , 0 y - t i k ,

(A l)

w here a ( k , 0 ) is the space-tim e Fourier am plitude o f the velocity field at wave num ber k and frequency 0 . The twopoint velocity correlation at frequency 0 (7) can be w ritten

VI. CONCLUSION In the present experim ent, the anisotropy o f the turbulent energy distribution at a given spatiotem poral scale ( r i , 0 ) is w ell described by the inertial-w ave dispersion relation at high frequency and/or large scale only. The sm aller-scale waves are subject to intense sw eeping by large-scale turbulent m otions contained predom inantly in the 2D vortex mode. This conclusion is com patible w ith the num erical findings o f Clark di Leoni et al. [38], who also identify the sw eeping tim e scale as the relevant decorrelation tim e at sm all scale. Such sw eeping by the 2D m ode has strong im plications for w ave-turbulence theories. Indeed, m ost waves do not follow the inertial-w ave dispersion relation and the assum ptions of w eak-turbulence theory break down even at the linear stage in wave am plitude: Instead o f the dispersion relation (1), the linear problem consists in determ ining the evolution o f waves

R ( r , 0 ) = ^ J J a (k , , 0 ) • a*(k 2, 0 ) X £ ,/(k1-x—kr (x+r))J k |c / k 2 + c c . j

=

J

|a ( k , 0 )|2 cos(k • r)t/k ,

(A 2 )

w here the angular brackets is the space average and r is a separation vector inside the PIV plane. Introducing spherical coordinates with vertical polar axis, we denote by